DiningRD

Revolutionizing Ingredient Import Efficiency with AI

Executive Summary:

DiningRD, a leading Dietitian Consulting and Foodservice Software Services provider serving over 4,000 healthcare communities across 46 U.S. states, partnered with ArchitectNow to overhaul its backend ingredients-import process. Facing significant efficiency challenges due to manual mapping of approximately 1,500 vendor-specific ingredients monthly, DiningRD implemented an AI-enabled solution to automate it. This automated search and match system, integrated with existing workflows, reduced manual effort, improved import efficiency, and positioned the company for future scalability with minimal disruption.

Challenges Faced:

DiningRD (DRD) struggled with the intensive manual task of mapping about 1,500 ingredients monthly, each with their own vendor specific unique naming convention, to a master ingredient list maintained by DRD. With approximately 30 different variations in names across various vendors that mapped to the same master ingredient, import administrators spent excessive hours on manually searching and matching one ingredient at a time. This bottleneck hindered productivity and delayed the integration of critical data into their software, impacting service delivery to long-term care communities.

Business Impact:

The manual process led to delayed recipe updates and increased operational costs, affecting supply chain vendors pas US as well as efficiencies of dietitian consultants. The lack of automation limited DiningRD's ability to scale this leg of its operation as vendor numbers and ingredient volumes grew. With approximately 250,000 vendor-specific ingredients mapped to about 8,000 master ingredients, the mapping process was already cumbersome.

Objectives:

The primary goal was to develop an AI-driven solution to automate the nightly import of vendor ingredient back-feeds, minimizing manual matching. The system needed to integrate seamlessly with existing dietitian-facing software, retain human oversight for matches below a set confidence threshold, and ensure scalability for future import volumes. The focus was on boosting productivity and efficiency with minimal changes to current systems.

The Solution:

Project Overview

ArchitectNow designed a tailored AI-enabled solution for DiningRD, leveraging Azure's AI Search and LLM based capabilities. The system automates the search and matching of vendor ingredients to the master list, processing imports nightly. A human-in-the-loop approach allows administrators to review and approve low-confidence matches, ensuring accuracy. The solution integrates with DiningRD's React-based frontend and existing SQL database, maintaining workflow continuity while enhancing scalability.

Key Features

• Automated Search and Match: AI identifies and maps vendor ingredients to the master list with high accuracy.

• Nightly Import: Processes 1,500 ingredients automatically each night, reducing manual workload.

• Human Oversight: The system flags matches with a confidence threshold below 90% for administrator review. This threshold is configurable to align with organizational requirements. Users can provide feedback by selecting a thumbs-up or thumbs-down on matches, enabling the AI search algorithm to learn and improve the accuracy of future matches.

• Scalable Architecture: Handles increased vendor and ingredient volumes without performance degradation.
• 
  
• Seamless Integration: Works within the existing React and SQL ecosystem, preserving user experience.

Technology Stack

The solution included the following technology components:

• Azure AI Search: Leverages indexer, synonym search, hybrid search, and scoring capabilities to enhance search functionality.
• Azure OpenAI Model: Employs a large language model to improve search performance for edge cases.
• Azure Storage Accounts Queue: Manages import jobs efficiently.
• Azure Functions: Implements the Microsoft Semantic Kernel framework, built with .NET, for streamlined data processing.
• Frontend: Utilizes React v18 with Redux Toolkit to deliver a robust and responsive user interface.
• SQL Database: Stores data securely and efficiently.
• Application Insights: Provides comprehensive log analytics for system monitoring.

Lessons Learnt

The project emphasized the importance of designing a system that balances AI automation with human oversight. This is achieved through a feature that enables users to set clear confidence thresholds for AI-generated matches.